Last wednesday, Meta released a completely new Model called Muse Spark. And according to their own data, it is kicking butt in the benchmarks. Out performing basically all of it’s competitors. So let’s dive in.

If you want to read more about other AI models check out this article on “The best AI Models in 2026”

Benchmarks

What we see above is a picture from Metas release article. It showcases how Muse spark is doing in some of the biggest benchmarks areas, compared to some of the biggest models out there. And it looks really good, but since this is released by Meta, we should always take this with a grain of salt.

Instead i did some testing myself, and i am honestly really impressed. I like the responses, it doesn’t feel programatic like i’ve always felt ChatGPT does. It reasons good, it solves problem, it creates nice pictures, and it does some good coding.

There is however some issues still. Especially in the agentic section where it is falling behind on benchmarks. Specifically the Terminal-Bench 2.0 and GDPval-AA Elo Elo. The number we see is a significant disadvantage, that will hopefully be closed a bit more, by the upcomming mode “Contemplating”.

Modes

Muse spark currently has 2 modes, with a third mode coming soon.

Instant mode :

Instant mode is you quick to answer, but not as deep thinking model. I would probably compare it to Claude Haiku, but i do believe it performs better than haiku.

Thinking mode :

Thinking mode is slower, I’ve asked it a couple of benchmark mathematical question and it’s been a long time in answering them. However, the difference in accuracy when going with these complex tasks is remarkable. I would compare this to Opus, but it’s hard to do so without having tried contemplating mode yet. For a full breakdown of Claude’s model tiers, see our Claude AI complete guide.

Contemplating mode :

Contemplating mode is not yet released, so i have not had the opportunity to try it yet. However they have released some benchmarks for it, and it looks extremely promising.

100 words sales pitch.

For this article, i decided to try something new that i might incorporate in my other deep dives. I asked Muse Spark to sell itself

Q : In a 100 words sales pitch, why should people use muse spark instead of other top tier LLM models?

A : Muse Spark blends top-tier reasoning with built-in creation. Unlike other LLMs, you can chat, search the live web, and generate or edit images and videos all in one flow—no switching tools. It’s conversational and playful, not stiff, yet still handles code, math, and deep analysis. You get up-to-date answers, seamless media generation with your likeness, and free access on meta.ai and the Meta AI app. For creators, researchers, and everyday users who want brains and a studio in one AI, Muse Spark delivers.

Pricing

Normally we would do a complete overview of the pricing of the model. With how much it costs, subscriptions, tokens and everything else. But the simple fact is, for now it is free.

You can use the model completly free of charge, but you can’t use the API key.

So how do we use it?

You can access it on Meta.ai, this is you classic chatbot that most of us have come to know and love.

Other than that, you can access it on all of Metas other products, Facebook, Instagram and so on.

Big question – Should you start using Muse spark?

Short answer – Yes.

Long answer – Why not? There is a lot on controversy going on right now with AI. There are plenty of conspiracies on Claude “nerfing” Opus 4.6, ChatGPT removing features, Claude lowering the limits on their Max and pro plan. Muse Spark is a good model, and it is free. So you really have nothing to lose if you want to test it out. So my advice is go for it.